Move our product from Docker Compose on one VM to EKS with zero-downtime deploys

Every deploy currently causes 40 seconds of downtime, because production is a single EC2 instance running Docker Compose with a pull-and-restart script. That was fine at 50 customers; at 900 it is embarrassing. We want the stack (three Node.js services, PostgreSQL on RDS, Redis) moved to EKS with rolling deployments and sane autoscaling. Deliverables: cluster setup in our AWS account (infrastructure as code, reviewed in PRs), Helm charts or Kustomize manifests for the services, a GitHub Actions pipeline that builds, pushes to ECR and deploys on merge, secrets handled through AWS Secrets Manager, and CloudWatch-based alerting for the obvious failure modes. We also want a runbook: what to do at 3am when a pod is crash-looping, written for engineers who are not Kubernetes experts. Our two backend developers should come out of this able to operate the cluster themselves, so plan for two or three knowledge-transfer sessions rather than disappearing with the keys.